AI Beat

GuideFor your jobHealth

AI for doctors: a practical guide

What ambient scribes and clinical AI tools actually deliver, what the evidence says, how to protect patient data, and which rules apply in the US, the UK and the EU.

A doctor holding a stethoscope
Photo: Alexandr Podvalny on Unsplash

In short

  • Four in five US physicians now use AI in their work, up from 38% in 2023, according to the American Medical Association.
  • Ambient AI scribes are the biggest change. In one multi-site study, burnout fell from 52% to 39% after a month of use; in a randomised trial the time saved per note was modest.
  • Scribes still make clinically significant mistakes, such as omissions and wrong pronouns. Every note needs your review before you sign it.
  • Patient data may only go into tools covered by an agreement with your organisation. In the US that means a business associate agreement; a consumer chatbot account is not HIPAA-compliant.
  • Patients should be told when a visit is being recorded by an AI scribe and must be able to say no.
  • Regulators draw the line at decisions: a tool that only transcribes and drafts is treated differently from one that suggests a diagnosis or treatment.

Medicine has moved from asking whether to use AI to asking how. Most doctors now use it in some form, usually to cut paperwork or look up evidence. This guide covers what the main tools do, what the research shows, and the privacy and safety rules that still apply when a machine is in the room.

How many doctors use AI

The American Medical Association has surveyed US physicians every year since 2023. In its 2026 survey, 81% said they use AI in their practice, more than double the 2023 figure. The most common uses are summarising medical research and clinical documentation.

Most US physicians now use AI at work

Share of physicians who say they use AI in their practice. The AMA did not publish a 2025 figure.

  1. 202338%
  2. 202466%
  3. 202681%

Source: American Medical Association physician surveys on augmented intelligence, 2023–2026

Enthusiasm comes with worry. In the same survey, 88% of doctors said they were concerned about losing skills, especially early in their careers, and 86% said data privacy is critical.

Ambient scribes: the evidence so far

An ambient scribe listens to the consultation, with the patient’s knowledge, and drafts the clinical note for you to review. The main products include Microsoft’s Dragon Copilot, Abridge and Nabla, and Epic now offers its own built into the health record.

The research is encouraging but more modest than the marketing:

  • Burnout. A study of 263 clinicians at six US health systems, published in JAMA Network Open, found burnout fell from 51.9% to 38.8% after 30 days with a scribe. After-hours documentation dropped by about 54 minutes a week.
  • Time per note. A randomised trial at UCLA, published in NEJM AI, compared two scribes with no scribe across 238 physicians. One tool cut note-writing time by about 41 seconds per note, roughly 10%. The other made no statistically significant difference. Both improved burnout scores by about 7%.
  • Accuracy. The UCLA trial found occasional clinically significant errors, such as omissions and wrong pronouns, and one mild patient-safety event.

The lesson: scribes make the job lighter, but the note is still yours. Read it before you sign it, and pay attention to what is missing as well as what is wrong.

Clinical question tools

A second group of tools answers clinical questions with references to the medical literature:

Tool Made by What it does Worth knowing
OpenEvidence OpenEvidence Answers clinical questions with citations to journals Aimed at verified US clinicians; reports over 200 million consultations to date
UpToDate Expert AI Wolters Kluwer Generative answers grounded in UpToDate’s reviewed content Sold through institutional subscriptions
ChatGPT for Healthcare OpenAI General AI assistant for health organisations Organisational plan with a business associate agreement; the consumer ChatGPT app is not covered by HIPAA
Claude for Healthcare Anthropic General AI assistant with medical data connectors HIPAA-ready enterprise offer; connectors include PubMed and ICD-10

Tools that cite sources make checking easier, but they do not remove the need to check. Open the key references before acting on an answer, particularly for dosing, rare conditions and recent guidance.

Protecting patient data

  • United States. Under HIPAA, a practice that sends patient information to a cloud service without a business associate agreement is breaking the rules. A personal chatbot account has no such agreement, so never paste identifiable patient details into one.
  • United Kingdom. NHS guidance on ambient scribes says patients must be told at the start that the tool is in use and must be able to object. Organisations must complete the usual data protection and clinical safety checks before deployment.
  • Everywhere. Use the tools your organisation has approved. If you experiment with anything else, strip out names, dates, record numbers and any detail that could identify a patient.

What the rules say

  • United States. The FDA regulates AI that works as a medical device, and its public list of AI-enabled devices now runs to more than a thousand products, most of them in imaging. Tools that only help with paperwork are generally outside its scope. States are adding their own rules: California requires a disclaimer and a way to reach a human when generative AI writes clinical messages to patients, unless a clinician has reviewed them, and Illinois bans AI from providing therapy or making therapeutic decisions.
  • United Kingdom. In July 2026 the MHRA clarified that ambient voice tools used only for transcription, summaries, letters or suggesting codes for a clinician to review are not medical devices. Tools that support diagnosis or treatment, or act without clinician review, are.
  • European Union. Under the AI Act, AI that is part of a regulated medical device counts as high-risk. After a 2026 amendment, those obligations apply from 2 August 2028.
  • Professional bodies. The AMA calls it “augmented intelligence”: AI should support doctors, not replace them, and patients and doctors should know when it is being used. The WHO’s 2024 guidance on large multimodal models sets out similar principles for health systems.

Five rules for safe use

  1. Review before you sign. Treat every AI-drafted note, letter or summary as a first draft.
  2. Keep decisions human. Use AI to gather and summarise, not to make the diagnosis or treatment decision.
  3. Tell patients. Explain when a scribe is recording and respect a patient who says no.
  4. Use approved tools only. Check that there is a contract, and in the US a business associate agreement, before any patient data goes in.
  5. Keep your skills sharp. Check the tool against your own judgement often, especially in training.

For the latest developments, follow the Health section, and look up unfamiliar terms in our AI glossary.

Comments

What do you think? No account needed. Please be kind; comments that attack people are removed.

We store only your name and comment. See our privacy policy.

    Sources

    1. American Medical Association: AI usage among doctors doubles as confidence in the technology grows (March 2026)
    2. American Medical Association: 2 in 3 physicians are using health AI, up 78% from 2023 (February 2025)
    3. American Medical Association: Augmented intelligence in medicine
    4. JAMA Network Open: Ambient AI scribes and clinician burnout, a multicenter quality improvement study (October 2025)
    5. UCLA Health: Randomised trial finds AI scribes may reduce documentation time (NEJM AI, November 2025)
    6. Microsoft: Dragon Copilot scales across care teams, partners and geographies (October 2025)
    7. OpenEvidence: About
    8. Wolters Kluwer: UpToDate Expert AI
    9. Anthropic: Claude for Healthcare and Life Sciences (January 2026)
    10. OpenAI: OpenAI for Healthcare (January 2026)
    11. US HHS: Using a cloud service provider without a business associate agreement
    12. US FDA: Artificial intelligence-enabled medical devices list
    13. NHS England Digital: Using AI-enabled ambient scribing products in health and care settings
    14. MHRA: MHRA clarifies regulatory status of ambient voice technologies used in the NHS (July 2026)
    15. European Commission: AI Omnibus enters into force (July 2026)
    16. California Legislature: AB 3030, health care services and artificial intelligence
    17. Illinois IDFPR: Governor signs HB 1806 restricting AI in therapy (August 2025)
    18. WHO: Ethics and governance guidance for large multi-modal models (January 2024)

    More guides